Arm CFO Eyes Deals as Chip Building Bring New Challenges

Arm shares have more than doubled this year as the U.K. company has benefited from demand for the central processing units it designs and a new AI chip it is selling. And that soaring share price has given the SoftBank-controlled company, now worth $300 billion, more firepower to do acquisitions.

In the past, Arm has mostly made small purchases, such as last year’s $265 million deal for networking startup DreamBig. It’s likely to stick to this kind of acquisition in the future, Arm Chief Financial Officer Jason Child said in an interview. But he indicated the company hasn’t ruled out bigger ones. That’s especially the case as its decision to sell its own chips, not just license its designs to companies like Nvidia and Apple, forces it to grapple with new growth and the challenges that come with it.

“Delivering silicon is definitely more complicated” than licensing designs, he said.

Arm announced in March that it would start selling CPUs for AI and had lined up Meta Platforms and OpenAI as customers. The company, which generated $4.9 billion in revenue in its last fiscal year, has said it expects more than $2 billion in customer demand for the new chips across fiscal 2027 and 2028. It’s one of many new entrants to developing AI chips, including Microsoft, Meta, OpenAI and Anthropic.

Child said the company’s timing to launch a specialized CPU for AI was fortuitous. The popularity of OpenClaw, open-source software for creating AI agents, has helped set off a frenzy for the personal assistants. These require cloud providers to draw on more Arm-designed CPUs, which use software tools to perform tasks over extended periods. These workloads rely heavily on memory and system orchestration of CPUs. Arm-designed CPUs also sit next to Nvidia’s graphics processing units in servers that developers use to train AI models.

The spike in CPU demand has boosted the fortunes of Intel, AMD, Nvidia and other CPU sellers but put further strains on underlying chip components, such as memory. And as a newcomer to selling chips, Arm can only command a relatively small amount of chipmaking capacity from factories like those run by Taiwan Semiconductor Manufacturing Co.

When it comes to selling its own AI chips, “the opportunity has gotten much larger than what we’d even planned when we started it,” he said. “The challenging part is now you’re seeing all the constraints.”

Child, a former Amazon executive who has been at Arm since 2022, also spoke about the semiconductor supply chain shortage, whether Arm could follow Intel in selling new equity to investors, and the relationship between the company and its parent, SoftBank.

This interview has been edited for length and clarity.

Arm is traditionally a capital-light business model, and the margins are very high, and now you’re selling finished chips. So where do you see new financial risks that Arm may need to take on?

When you’re trying to build chips, you have to kind of get in line. And the way they generally allocate capacity is, what did you get last year, and then I’m going to give you some increase as I increase my capacity.

And so if you’re starting from zero, that means you have a pretty small amount of capacity. And so it’s going to take a couple years to ramp up your capacity. It’s gonna take a couple years to get our share of the fabrication capacity, the memory capacity, and so on. That’s one of the things that has been a challenge.

The other part that’s been a challenge is because of how quickly prices have increased: Demand is so far exceeding supply. That basically means it’s really hard to be in this business unless you have significant amounts of capital. If you look at who is putting all the money into everything right now, it’s mostly the hyperscalers. They fund it with free cash flow because they all have incredible business models.

What that means is, if you’re a smaller company and you’re trying to go and enter in this market, it’s really hard, and [that’s] why a company like Groq sold to Nvidia.

You still have to come up with a lot of capital and a lot of borrowing capacity to go build out all of the capex necessary to build out the chip volume necessary. That’s why we set the financial expectations somewhat conservative, and basically gave ourselves a couple years to grow into it.

What are your views on the CPU crunch?

The need for compute keeps growing. The challenges probably are mostly around how much power can all the different folks who are investing all the money in capex [secure]. A chip is just one piece. You need to have a full system. You need to have a data center to house all those systems, and of course you need power to then produce the tokens.

As far as I can tell, the demand on compute is for the most part insatiable. It’s about having to figure out: I want to buy the chip, but then I also have to figure out how to get memory, which is highly constrained. I have to figure out how to make sure that these chips can be produced and fabricated, and so then I’ve got to go make sure there’s fabrication capacity through mostly TSMC today, and then I also, of course, then have to figure out if I have power.

So we’re just one of those pieces, but as far as I can tell, there’s no slowdown at all.

Where do you feel like the most important supply bottleneck is today, and what can Arm do about it?

Memory is high on the list. That said, TSMC has also talked about wafer capacity and throughput capacity because they only have so many fabs that are actually up and running and producing at the nanometer size that companies need.

You hear some of the model companies or some of the hyperscalers talking about trying to stand up 1 to 5 gigawatts [of data centers.] Those are pretty significant amounts of power. And then you get into the questions about these data centers: Where are they going to be? Can you get power to support them? Everyone sees it being highly constrained in the next couple years.

How do you expect Arm’s investment priorities to change? The company now takes more responsibility for the chip development process as it moves from just licensing IPs.

When I think about all the stuff we’re trying to build, you first think about what can we build from within and then what do we actually have to go acquire.

So on M&A, I then would break it down between what is going to help us expand our total addressable market, and then second, what is the capability to increase our ability to do more on our own, which will actually help us improve margins.

In the semiconductor space, it’s famous that no one builds anything completely end to end on their own—no one. Nvidia is relatively close, and maybe Apple’s relatively [close]. And so that’s how folks like Broadcom or even Nvidia or Apple have been able to have such high margins in their business because they own so much of the stack.

That said, I would say we’re still pretty early in our journey, and so as a result, a lot of the M&A we’ve done thus far has been mostly more about TAM [total addressable market] expansion. The most recent company we bought was DreamBig. That’s a networking business. We don’t really have any real deep capabilities in networking, so that’s really just helping us to be able to expand our TAM.

For the most part, we’ve focused on smaller companies—ones that are probably private. The company’s acquired about 20 companies over the past 15, 20 years, and it’s mostly been through that type of approach. It’s worked pretty well for us. We haven’t done much large-scale public company M&A.

Of course, we will keep our ears and eyes open and look at stuff.

It was reported that you made an eleventh-hour bid for Cerebras and considered a huge merger with Marvell Technology last year. What would you consider for future acquisitions?

Certainly we will evaluate everything. Sometimes there’s things that come to us, sometimes there’s things that we want to just kind of go look at. For the most part, we’re going to look at everything.

When I look at the things out there, large M&A is certainly [a] capability. We are 87% owned by SoftBank. Masa [Masayoshi Son, SoftBank CEO ] is our chairman, and if you look at what he’s done, he’s acquired Graphcore: When SoftBank purchased Graphcore, they were one of the many companies focused on an XPU [specialized processor unit], as they called it, and there was Graphcore, there was SambaNova, there was Cerebras, there was Groq, and I know all those companies were evaluated.

Arm always has this close relationship with SoftBank, and now Arm CEO Rene Haas is also CEO of SoftBank International. How do you coordinate those with the SoftBank universe of companies that could complement Arm’s work?

It made sense for Rene to take a role as CEO of SoftBank Group International. It’s an operating group that basically owns the assets within Masa’s organization that is focused on delivering ASI [artificial superintelligence], which is his grand vision. Everything Rene is in charge of is mostly as it relates to helping develop chips and systems to be able to fulfill some of the demand Masa announced for Stargate.

Masa has the ultimate vision, and he’s the one that really is in charge of how capital should be allocated, the strategy and the plans for the company. Rene is in charge of the execution.

It’s been really helpful. It just helps us move a little faster because then you know when you have someone at SoftBank who’s in charge of these efforts, instead of going through multiple layers of approval, he can talk with Masa.

As you’ve probably noticed, the world of AI moves at a speed that seems to be beyond anything we’ve ever seen before, which is exciting but puts stress on the system. And so you really have to come up with unique operating structures.

Even the most profitable kind of hyperscalers have to take advantage of their growing stock prices to issue equities. Do you feel like we should expect something like this from other chipmakers or even Arm?

Nothing to announce today. But if you step back, when the markets are really pulling back, that’s typically when you see people do buybacks. And when stocks are at all-time highs—or relatively high valuations—that’s typically the time that you see folks issue more equity. I would say those are certainly things to consider.

Because of our relationship with SoftBank, they probably are a little more experienced and more advanced when it comes to some of the financial and capital aspects, and so for the most part, we’ve partnered with them on a number of more riskier or really capital-intensive projects by basically selling them a license and providing design services for them and building something for them.

They’re now a really big customer for some AI-oriented products that they’ll be talking about sometime down the road, and so that’s one way that’s prevented us from having to go out and issue a bunch of debt. It’d be really hard to do without a partnership with someone like SoftBank.

There are so many financing partnerships that chipmakers strike with Wall Street firms, and that entails more interesting structures that otherwise we may not see. And it’s getting more creative. What are your overall observations from this?

There’s a level of capital being deployed like we’ve never seen. I was at Amazon in ’99, 2000, 2001, and we went through that whole kind of crazy build-out. At that point, it was fascinating because there was the promise of e-commerce, but the actual experience of e-commerce was still pretty crappy. It took you a week or two to get your product. You had all these companies that were getting funded, but they weren’t using much of the capacity. It took 15 or 20 years to utilize all of the hardware and all of the capex that was built out.

This one’s very interesting because it’s all getting deployed immediately. It’s really more about what are the evolving business models and what’s the ROI [return on investment] going to look like for all the different businesses.

For the most part, if you look at where all the capital is really being put, it’s mostly being supported by businesses that have very significant cash flows—whether it’s Amazon or Meta or Google or Microsoft. Even if everything fell apart, they would still in the worst-case scenario [to] take a big write-off, and the businesses would still be fine.